87081c810f60fe87de916a3e0aa95333ea96a223
[platform/upstream/ltrace.git] / dwarf_prototypes.c
1 /* Most of this is Copyright Dima Kogan <dima@secretsauce.net>
2  *
3  * Pieces of this were taken from dwarf_prototypes.c in the dwarves project.
4  * Those are Copyright (C) 2008 Arnaldo Carvalho de Melo <acme@redhat.com>.
5  *
6  * This program is free software; you can redistribute it and/or modify it under
7  * the terms of version 2 of the GNU General Public License as published by the
8  * Free Software Foundation.
9  *
10  */
11 #include <stdio.h>
12 #include <elfutils/libdwfl.h>
13 #include <dwarf.h>
14 #include <stdlib.h>
15 #include <errno.h>
16 #include <string.h>
17
18 #include "config.h"
19 #include "prototype.h"
20 #include "type.h"
21 #include "param.h"
22 #include "dict.h"
23 #include "lens.h"
24 #include "lens_enum.h"
25 #include "value.h"
26 #include "expr.h"
27 #include "library.h"
28 #include "options.h"
29 #include "filter.h"
30
31
32 //#define DUMP_PROTOTYPES
33
34 #if 1
35 #define complain( die, format, ... )                                                    \
36         fprintf(stderr, "%s() die '%s' @ 0x%lx: " format "\n",          \
37                         __func__, dwarf_diename(die), dwarf_dieoffset(die),     \
38                         ##__VA_ARGS__ )
39 #else
40 #define complain( die, format, ... )
41 #endif
42
43 // A map from DIE addresses (Dwarf_Off) to type structures (struct
44 // arg_type_info*). This is created and filled in at the start of each import,
45 // and deleted when the import is complete
46 static struct dict type_hash;
47
48
49 static bool getType( struct arg_type_info** info, Dwarf_Die* type_die);
50
51
52 #if 0
53 static bool _dump_dwarf_tree(Dwarf_Die* die, int indent)
54 {
55     while(1)
56     {
57         fprintf(stderr, "%*sprocessing unit: 0x%02x/'%s'\n", indent*4, "",
58                dwarf_tag(die), dwarf_diename(die) );
59
60         Dwarf_Die child;
61         if (dwarf_child(die, &child) == 0)
62         {
63                         if( !_dump_dwarf_tree(&child, indent+1) )
64                                 return false;
65         }
66
67         int res = dwarf_siblingof(die, die);
68         if( res == 0 ) continue;     // sibling exists
69         if( res < 0 )  return false; // error
70         break;                       // no sibling exists
71     }
72
73     return true;
74 }
75
76 static bool dump_dwarf_tree(Dwarf_Die* die)
77 {
78     return _dump_dwarf_tree( die, 0 );
79 }
80 #endif
81
82 #ifdef DUMP_PROTOTYPES
83 static bool _dump_ltrace_tree( const struct arg_type_info* info, int indent )
84 {
85         if( indent > 7 )
86         {
87                 fprintf(stderr, "%*s%p ...\n", indent*4, "", (void*)info);
88                 return true;
89         }
90
91         if( info == NULL )
92         {
93                 fprintf(stderr, "%*s%p NULL\n", indent*4, "", (void*)info);
94                 return true;
95         }
96
97         switch(info->type)
98         {
99         case ARGTYPE_VOID:
100                 fprintf(stderr, "%*s%p void\n", indent*4, "", (void*)info);
101                 break;
102
103         case ARGTYPE_INT:
104         case ARGTYPE_UINT:
105         case ARGTYPE_LONG:
106         case ARGTYPE_ULONG:
107         case ARGTYPE_CHAR:
108         case ARGTYPE_SHORT:
109         case ARGTYPE_USHORT:
110         case ARGTYPE_FLOAT:
111         case ARGTYPE_DOUBLE:
112                 fprintf(stderr, "%*s%p base\n", indent*4, "", (void*)info);
113                 break;
114
115         case ARGTYPE_ARRAY:
116                 fprintf(stderr, "%*s%p array. elements not printed\n", indent*4, "", (void*)info);
117                 break;
118
119         case ARGTYPE_POINTER:
120                 fprintf(stderr, "%*s%p pointer to...\n", indent*4, "", (void*)info);
121                 _dump_ltrace_tree( info->u.ptr_info.info, indent+1 );
122                 break;
123
124         case ARGTYPE_STRUCT:
125                 fprintf(stderr, "%*s%p struct...\n", indent*4, "", (void*)info);
126                 struct struct_field
127                 {
128                         struct arg_type_info *info;
129                         int own_info;
130                 }* elements = (struct struct_field*)info->u.entries.data;
131                 unsigned int i;
132                 for(i=0; i<info->u.entries.size; i++)
133                         _dump_ltrace_tree( elements[i].info, indent+1 );
134                 break;
135
136         default:
137                 fprintf(stderr, "%*s%p unknown type\n", indent*4, "", (void*)info);
138                 return false;;
139         }
140
141         return true;
142 }
143
144 static bool dump_ltrace_tree( const struct arg_type_info* info )
145 {
146         return _dump_ltrace_tree( info, 0 );
147 }
148 #endif
149
150
151
152 static uint64_t attr_numeric(Dwarf_Die *die, uint32_t name)
153 {
154         Dwarf_Attribute attr;
155         uint32_t form;
156
157         if (dwarf_attr(die, name, &attr) == NULL)
158                 return 0;
159
160         form = dwarf_whatform(&attr);
161
162         switch (form) {
163         case DW_FORM_addr: {
164                 Dwarf_Addr addr;
165                 if (dwarf_formaddr(&attr, &addr) == 0)
166                         return addr;
167         }
168                 break;
169         case DW_FORM_data1:
170         case DW_FORM_data2:
171         case DW_FORM_data4:
172         case DW_FORM_data8:
173         case DW_FORM_sdata:
174         case DW_FORM_udata: {
175                 Dwarf_Word value;
176                 if (dwarf_formudata(&attr, &value) == 0)
177                         return value;
178         }
179                 break;
180         case DW_FORM_flag:
181         case DW_FORM_flag_present: {
182                 bool value;
183                 if (dwarf_formflag(&attr, &value) == 0)
184                         return value;
185         }
186                 break;
187         default:
188                 complain(die, "DW_AT_<0x%x>=0x%x", name, form);
189                 break;
190         }
191
192         return 0;
193 }
194
195 static enum arg_type getBaseType( Dwarf_Die* die )
196 {
197         int encoding = attr_numeric(die, DW_AT_encoding);
198
199         if( encoding == DW_ATE_void )
200                 return ARGTYPE_VOID;
201
202         if( encoding == DW_ATE_signed_char || encoding == DW_ATE_unsigned_char )
203                 return ARGTYPE_CHAR;
204
205         if( encoding == DW_ATE_signed ||
206                 encoding == DW_ATE_unsigned )
207         {
208                 bool is_signed = (encoding == DW_ATE_signed);
209                 switch( attr_numeric(die, DW_AT_byte_size) )
210                 {
211                 case sizeof(char):
212                         return ARGTYPE_CHAR;
213
214                 case sizeof(short):
215                         return is_signed ? ARGTYPE_SHORT : ARGTYPE_USHORT;
216
217                 case sizeof(int):
218                         return is_signed ? ARGTYPE_INT : ARGTYPE_UINT;
219
220                 case sizeof(long):
221                         return is_signed ? ARGTYPE_LONG : ARGTYPE_ULONG;
222
223                 default:
224                         complain(die, "");
225                         exit(1);
226                 }
227         }
228
229         if( encoding == DW_ATE_float )
230         {
231                 switch( attr_numeric(die, DW_AT_byte_size) )
232                 {
233                 case sizeof(float):
234                         return ARGTYPE_FLOAT;
235
236                 case sizeof(double):
237                         return ARGTYPE_DOUBLE;
238
239                 default:
240                         complain(die, "");
241                         exit(1);
242                 }
243         }
244
245         complain(die, "");
246         exit(1);
247         return ARGTYPE_VOID;
248 }
249
250 static bool getTypeDie( Dwarf_Die* type_die, Dwarf_Die* die )
251 {
252         Dwarf_Attribute attr;
253         return
254                 dwarf_attr(die, DW_AT_type, &attr) != NULL &&
255                 dwarf_formref_die(&attr, type_die) != NULL;
256 }
257
258 static size_t dwarf_die_hash(const void* x)
259 {
260         return *(const Dwarf_Off*)x;
261 }
262 static int dwarf_die_eq(const void* a, const void* b)
263 {
264         return *(const Dwarf_Off*)a == *(const Dwarf_Off*)b;
265 }
266
267 static bool getEnum(struct arg_type_info* enum_info, Dwarf_Die* parent)
268 {
269         enum_info->type = ARGTYPE_INT;
270
271         struct enum_lens *lens = calloc(1, sizeof(struct enum_lens));
272         if (lens == NULL)
273         {
274                 complain(parent, "alloc error");
275                 return false;
276         }
277         lens_init_enum(lens);
278         enum_info->lens = &lens->super;
279
280         Dwarf_Die die;
281         if( dwarf_child(parent, &die) != 0 )
282         {
283                 // empty enum. we're done
284                 return true;
285         }
286
287         while(1) {
288                 complain(&die, "enum element: 0x%02x/'%s'", dwarf_tag(&die), dwarf_diename(&die) );
289
290                 if( dwarf_tag(&die) != DW_TAG_enumerator )
291                 {
292                         complain(&die, "Enums can have ONLY DW_TAG_enumerator elements");
293                         return false;
294                 }
295
296                 if( !dwarf_hasattr(&die, DW_AT_const_value) )
297                 {
298                         complain(&die, "Enums MUST have DW_AT_const_value values");
299                         return false;
300                 }
301
302                 const char* key = dwarf_diename(&die);
303                 if( key == NULL )
304                 {
305                         complain(&die, "Enums must have a DW_AT_name key");
306                         return false;
307                 }
308                 const char* dupkey = strdup(key);
309                 if( dupkey == NULL )
310                 {
311                         complain(&die, "Couldn't duplicate enum key");
312                         return false;
313                 }
314
315                 struct value* value = calloc( 1, sizeof(struct value) );
316                 if( value == NULL )
317                 {
318                         complain(&die, "Couldn't alloc enum value");
319                         return false;
320                 }
321
322                 value_init_detached(value, NULL, type_get_simple( ARGTYPE_INT ), 0);
323                 value_set_word(value, attr_numeric(&die, DW_AT_const_value) );
324
325                 if( lens_enum_add( lens, dupkey, 0, value, 0 ) )
326                 {
327                         complain(&die, "Couldn't add enum element");
328                         return false;
329                 }
330
331                 int res = dwarf_siblingof(&die, &die);
332                 if( res == 0 ) continue;     /* sibling exists    */
333                 if( res < 0 )  return false; /* error             */
334                 break;                       /* no sibling exists */
335         }
336
337         return true;
338 }
339
340 static bool getArray(struct arg_type_info* array_info, Dwarf_Die* parent)
341 {
342         Dwarf_Die type_die;
343         if( !getTypeDie( &type_die, parent ) )
344         {
345                 complain( parent, "Array has unknown type" );
346                 return false;
347         }
348
349         struct arg_type_info* info;
350         if( !getType( &info, &type_die ) )
351         {
352                 complain( parent, "Couldn't figure out array's type" );
353                 return false;
354         }
355
356         Dwarf_Die subrange;
357         if( dwarf_child(parent, &subrange) != 0 )
358         {
359                 complain( parent, "Array must have a DW_TAG_subrange_type child, but has none" );
360                 return false;
361         }
362
363         Dwarf_Die next_subrange;
364         if( dwarf_siblingof(&subrange, &next_subrange) <= 0 )
365         {
366                 complain( parent, "Array must have exactly one DW_TAG_subrange_type child" );
367                 return false;
368         }
369
370         if( !dwarf_hasattr(&subrange, DW_AT_upper_bound) )
371         {
372                 complain( parent, "Array subrange must have a DW_AT_upper_bound");
373                 return false;
374         }
375
376         if( dwarf_hasattr(&subrange, DW_AT_lower_bound) )
377         {
378                 if( attr_numeric(&subrange, DW_AT_lower_bound) != 0 )
379                 {
380                         complain( parent, "Array subrange has a nonzero lower bound. Don't know what to do");
381                         return false;
382                 }
383         }
384
385         // I'm not checking the subrange type. It should be some sort of integer,
386         // and I don't know what it would mean for it to be something else
387
388         struct value* value = calloc( 1, sizeof(struct value) );
389         if( value == NULL )
390         {
391                 complain(&subrange, "Couldn't alloc length value");
392                 return false;
393         }
394         value_init_detached(value, NULL, type_get_simple( ARGTYPE_INT ), 0);
395         value_set_word(value, attr_numeric(&subrange, DW_AT_upper_bound)+1 );
396
397         struct expr_node* length = calloc( 1, sizeof(struct expr_node) );
398         if( length == NULL )
399         {
400                 complain(&subrange, "Couldn't alloc length expr");
401                 return false;
402         }
403         expr_init_const(length, value);
404
405         type_init_array(array_info, info, 0, length, 0 );
406
407         return true;
408 }
409
410 static bool getStructure(struct arg_type_info* struct_info, Dwarf_Die* parent)
411 {
412         type_init_struct(struct_info);
413
414         Dwarf_Die die;
415         if( dwarf_child(parent, &die) != 0 )
416         {
417                 // no elements; we're done
418                 return true;
419         }
420
421         while(1) {
422                 fprintf(stderr, "member: 0x%02x/'%s'\n", dwarf_tag(&die), dwarf_diename(&die) );
423
424                 if( dwarf_tag(&die) != DW_TAG_member )
425                 {
426                         complain(&die, "Structure can have ONLY DW_TAG_member");
427                         return false;
428                 }
429
430                 Dwarf_Die type_die;
431                 if( !getTypeDie( &type_die, &die ) )
432                 {
433                         complain( &die, "Couldn't get type of element");
434                         return false;
435                 }
436
437                 struct arg_type_info* member_info = NULL;
438                 if( !getType( &member_info, &type_die ) )
439                 {
440                         complain(&die, "Couldn't parse type from DWARF data");
441                         return false;
442                 }
443                 type_struct_add( struct_info, member_info, 0 );
444
445                 int res = dwarf_siblingof(&die, &die);
446                 if( res == 0 ) continue;     /* sibling exists    */
447                 if( res < 0 )  return false; /* error             */
448                 break;                       /* no sibling exists */
449         }
450
451         return true;
452 }
453
454 // Reads the type in the die into the given structure
455 // Returns true on sucess
456 static bool getType( struct arg_type_info** info, Dwarf_Die* type_die)
457 {
458         Dwarf_Off die_offset = dwarf_dieoffset(type_die);
459         struct arg_type_info** found_type = dict_find(&type_hash, &die_offset );
460         if(found_type != NULL)
461         {
462                 *info = *found_type;
463                 complain(type_die, "Read pre-computed type: %p", *info);
464                 return true;
465         }
466
467         Dwarf_Die next_die;
468
469         switch( dwarf_tag(type_die) )
470         {
471         case DW_TAG_base_type:
472                 *info = type_get_simple( getBaseType( type_die ) );
473                 complain(type_die, "Storing base type: %p", *info);
474                 dict_insert( &type_hash, &die_offset, info );
475                 return true;
476
477         case DW_TAG_subroutine_type:
478         case DW_TAG_inlined_subroutine:
479                 // function pointers are stored as void*. If ltrace tries to dereference
480                 // these, it'll get a segfault
481                 *info = type_get_simple( ARGTYPE_VOID );
482                 complain(type_die, "Storing subroutine type: %p", *info);
483                 dict_insert( &type_hash, &die_offset, info );
484                 return true;
485
486         case DW_TAG_pointer_type:
487
488                 if( !getTypeDie(&next_die, type_die ) )
489                 {
490                         // the pointed-to type isn't defined, so I report a void*
491                         *info = type_get_simple( ARGTYPE_VOID );
492                         complain(type_die, "Storing void-pointer type: %p", *info);
493                         dict_insert( &type_hash, &die_offset, info );
494                         return true;
495                 }
496
497                 *info = calloc( 1, sizeof(struct arg_type_info) );
498                 if( *info == NULL )
499                 {
500                         complain(type_die, "alloc error");
501                         return false;
502                 }
503                 type_init_pointer(*info, NULL, 0);
504
505                 complain(type_die, "Storing pointer type: %p", *info);
506                 dict_insert( &type_hash, &die_offset, info );
507                 return getType( &(*info)->u.ptr_info.info, &next_die );
508
509         case DW_TAG_structure_type:
510                 *info = calloc( 1, sizeof(struct arg_type_info) );
511                 if( *info == NULL )
512                 {
513                         complain(type_die, "alloc error");
514                         return false;
515                 }
516
517                 complain(type_die, "Storing struct type: %p", *info);
518                 dict_insert( &type_hash, &die_offset, info );
519                 return getStructure( *info, type_die );
520
521
522         case DW_TAG_typedef: ;
523         case DW_TAG_const_type: ;
524         case DW_TAG_volatile_type: ;
525                 // Various tags are simply pass-through, so I just keep going
526                 bool res = true;
527                 if( getTypeDie(&next_die, type_die ) )
528                 {
529                         complain(type_die, "Storing const/typedef type: %p", *info);
530                         res = getType( info, &next_die );
531                 }
532                 else
533                 {
534                         // no type. Use 'void'. Normally I'd think this is bogus, but stdio
535                         // typedefs something to void
536                         *info = type_get_simple( ARGTYPE_VOID );
537                         complain(type_die, "Storing void type: %p", *info);
538                 }
539                 if( res )
540                         dict_insert( &type_hash, &die_offset, info );
541                 return res;
542
543         case DW_TAG_enumeration_type:
544                 // We have an enumeration. This has type "int", but has a particular
545                 // lens to handle the enum
546                 *info = calloc( 1, sizeof(struct arg_type_info) );
547                 if( *info == NULL )
548                 {
549                         complain(type_die, "alloc error");
550                         return false;
551                 }
552
553                 complain(type_die, "Storing enum int: %p", *info);
554                 dict_insert( &type_hash, &die_offset, info );
555                 return getEnum( *info, type_die );
556
557         case DW_TAG_array_type:
558                 *info = calloc( 1, sizeof(struct arg_type_info) );
559                 if( *info == NULL )
560                 {
561                         complain(type_die, "alloc error");
562                         return false;
563                 }
564
565                 complain(type_die, "Storing array: %p", *info);
566                 dict_insert( &type_hash, &die_offset, info );
567                 return getArray( *info, type_die );
568
569         default:
570                 complain(type_die, "Unknown type tag 0x%x", dwarf_tag(type_die));
571                 break;
572         }
573
574         return false;
575 }
576
577 static bool getPrototype(struct prototype* proto, Dwarf_Die* subroutine)
578 {
579         // First, look at the return type. This is stored in a DW_AT_type tag in the
580         // subroutine DIE. If there is no such tag, this function returns void
581         Dwarf_Die return_type_die;
582         if( !getTypeDie(&return_type_die, subroutine ) )
583         {
584                 proto->return_info = type_get_simple( ARGTYPE_VOID );
585                 proto->own_return_info = 0;
586         }
587         else
588         {
589                 proto->return_info = calloc( 1, sizeof( struct arg_type_info ) );
590                 if( proto->return_info == NULL )
591                 {
592                         complain(subroutine, "Couldn't alloc return type");
593                         return false;
594                 }
595                 proto->own_return_info = 0;
596
597                 if( !getType( &proto->return_info, &return_type_die ) )
598                 {
599                         complain(subroutine, "Couldn't get return type");
600                         return false;
601                 }
602         }
603
604
605         // Now look at the arguments
606         Dwarf_Die arg_die;
607         if( dwarf_child(subroutine, &arg_die) != 0 )
608         {
609                 // no args. We're done
610                 return true;
611         }
612
613         while(1) {
614                 if( dwarf_tag(&arg_die) != DW_TAG_formal_parameter )
615                         goto next_prototype_argument;
616
617                 complain(&arg_die, "arg: 0x%02x", dwarf_tag(&arg_die));
618
619                 Dwarf_Die type_die;
620                 if( !getTypeDie(&type_die, &arg_die ) )
621                 {
622                         complain(&arg_die, "Couldn't get the argument type die");
623                         return false;
624                 }
625
626                 struct arg_type_info* arg_type_info = NULL;
627                 if( !getType( &arg_type_info, &type_die ) )
628                 {
629                         complain(&arg_die, "Couldn't parse arg type from DWARF data");
630                         return false;
631                 }
632
633                 struct param param;
634                 param_init_type(&param, arg_type_info, 0);
635                 if( prototype_push_param(proto, &param) <0 )
636                 {
637                         complain(&arg_die, "couldn't add argument to the prototype");
638                         return false;
639                 }
640
641 #ifdef DUMP_PROTOTYPES
642                 fprintf(stderr, "Adding argument:\n");
643                 dump_ltrace_tree(arg_type_info);
644 #endif
645
646         next_prototype_argument: ;
647                 int res = dwarf_siblingof(&arg_die, &arg_die);
648                 if( res == 0 ) continue;     /* sibling exists    */
649                 if( res < 0 )  return false; /* error             */
650                 break;                       /* no sibling exists */
651         }
652
653         return true;
654 }
655
656 static bool process_die_compileunit(struct protolib* plib, struct library* lib, Dwarf_Die* parent)
657 {
658         Dwarf_Die die;
659         if( dwarf_child(parent, &die) != 0 )
660                 return false;
661
662         while(1)
663         {
664                 if( dwarf_tag(&die) == DW_TAG_subprogram )
665                 {
666                         const char* function_name = dwarf_diename(&die);
667
668                         complain(&die, "subroutine_type: 0x%02x; function '%s'", dwarf_tag(&die), function_name);
669
670                         struct prototype* proto =
671                                 protolib_lookup_prototype(plib, function_name, true );
672
673                         if( proto != NULL )
674                         {
675                                 complain(&die, "Prototype already exists. Skipping");
676                                 goto next_prototype;
677                         }
678
679                         if( !filter_matches_symbol(options.plt_filter,    function_name, lib) &&
680                                 !filter_matches_symbol(options.static_filter, function_name, lib) &&
681                                 !filter_matches_symbol(options.export_filter, function_name, lib) )
682                         {
683                                 complain(&die, "Prototype not requested by any filter");
684                                 goto next_prototype;
685                         }
686
687                         proto = malloc(sizeof(struct prototype));
688                         if( proto == NULL )
689                         {
690                                 complain(&die, "couldn't alloc prototype");
691                                 return false;
692                         }
693                         prototype_init( proto );
694
695                         if( !getPrototype(proto, &die ) )
696                         {
697                                 complain(&die, "couldn't get prototype");
698                                 return false;
699                         }
700
701                         protolib_add_prototype(plib, function_name, 0, proto);
702                 }
703
704                 next_prototype:;
705                 int res = dwarf_siblingof(&die, &die);
706                 if( res == 0 ) continue;     /* sibling exists    */
707                 if( res < 0 )  return false; /* error             */
708                 break;                       /* no sibling exists */
709         }
710
711         return true;
712 }
713
714 static bool import( struct protolib* plib, struct library* lib, Dwfl* dwfl )
715 {
716         dict_init(&type_hash, sizeof(Dwarf_Off), sizeof(struct arg_type_info*),
717                           dwarf_die_hash, dwarf_die_eq, NULL );
718
719         Dwarf_Addr bias;
720     Dwarf_Die* die = NULL;
721     while( (die = dwfl_nextcu(dwfl, die, &bias)) != NULL )
722     {
723         if( dwarf_tag(die) == DW_TAG_compile_unit )
724         {
725             if( !process_die_compileunit(plib, lib, die) )
726             {
727                 complain(die, "Error reading compile unit");
728                                 exit(1);
729                                 return false;
730             }
731         }
732         else
733         {
734             complain(die, "DW_TAG_compile_unit expected");
735                         exit(1);
736             return false;
737         }
738     }
739
740         dict_destroy( &type_hash, NULL, NULL, NULL );
741         return true;
742 }
743
744 bool import_DWARF_prototypes( struct protolib* plib, struct library* lib,
745                                                           Dwfl *dwfl )
746 {
747         if( plib == NULL )
748         {
749                 plib = protolib_cache_default(&g_protocache, lib->soname, 0);
750                 if (plib == NULL)
751                 {
752                         fprintf(stderr, "Error loading protolib %s: %s.\n",
753                                         lib->soname, strerror(errno));
754                 }
755         }
756
757         return import(plib, lib, dwfl);
758 }
759
760 /*
761 - I handle static functions now. Should I? Those do not have DW_AT_external==1
762
763 - should process existing prototypes to make sure they match
764
765 - what do function pointers look like? I'm doing void*
766
767 - unions
768
769 - all my *allocs leak
770
771 - protolib_lookup_prototype should look for imports?
772
773 */